Baptiste Rozière (@b

Baptiste Rozière

186 posts

Baptiste Rozière

@b_roziere

CodeGen lead at Mistral AI. Devstral/Codestral/Codestral Embed

Paris, France

Joined September 2016

Pinned
Baptiste Rozière
@b_roziere
Aug 24, 2023
Today, we release CodeLlama, a collection of base and instruct-finetuned models with 7B, 13B and 34B parameters. For coding tasks, CodeLlama 7B is competitive with Llama 2 70B and CodeLlama 34B is state-of-the-art among open models. Paper and weights: ai.meta.com/research/publi…
213K
Baptiste Rozière
@b_roziere
Oct 15, 2025
We’re looking for talented interns to join our CodeGen team at mistral in Paris & Palo Alto. Preferably PhD or Master’s students planning to graduate in 2026 or early 2027. If you want to work on improving our LLM models for code generation and agents, please apply and DM me!
34K
Baptiste Rozière
@b_roziere
Sep 13, 2024
I'm thrilled to announce that I've recently joined @MistralAI! While I'll miss my former colleagues at Meta, I'm excited to continue building models for code generation with the incredible team here.
46K
Baptiste Rozière
@b_roziere
Aug 26, 2023
It's nice to see how quickly the open-source community was able to build on CodeLlama to surpass GPT4 on HumanEval!
WizardLM
@WizardLM_AI
Aug 26, 2023
🔥🔥🔥 Introduce the newest WizardCoder 34B based on Code Llama. ✅WizardCoder-34B surpasses GPT-4, ChatGPT-3.5 and Claude-2 on HumanEval with 73.2% pass@1 🖥️Demo: http://47.103.63.15:50085/ 🏇Model Weights: huggingface.co/WizardLM/Wizar… 🏇Github: github.com/nlpxucan/Wizar… The 13B/7B
71K
Baptiste Rozière
@b_roziere
Jan 29, 2024
We released a 70B version of CodeLlama today! Trained on 1T tokens, it is a much stronger base model for coding tasks. I look forward to seeing what the community will do with it! :)
AI at Meta
@AIatMeta
Jan 29, 2024
Today we’re releasing Code Llama 70B: a new, more performant version of our LLM for code generation — available under the same license as previous Code Llama models. Download the models ➡️ bit.ly/3Oil6bQ • CodeLlama-70B • CodeLlama-70B-Python • CodeLlama-70B-Instruct
29K
Baptiste Rozière
@b_roziere
May 21, 2025
We released Devstral. It is a 24B model released under the Apache 2.0 license. It the best open model on SWE-Bench verified today. You can check our blog post or test it with OpenHands (from @allhands_ai ) following the instructions here:
Mistral AI
@MistralAI
May 21, 2025
Meet Devstral, our SOTA open model designed specifically for coding agents and developed with @allhands_ai mistral.ai/news/devstral
mistralai/Devstral-Small-2505 · Hugging Face
From huggingface.co
15K
Baptiste Rozière
@b_roziere
Aug 24, 2023
Replying to @b_roziere
All the models we release are fine-tuned on sequences of 16k tokens with an increased RoPE theta parameter, allowing them to gain long-sequence processing capabilities at low cost. CodeLlama can extrapolate, and we show improvements for inputs of up to 100K tokens.
24K
Baptiste Rozière
@b_roziere
Feb 16, 2021
Replying to @nicolaskruchten and @GuillaumeLample
You can. To test our model, we actually developed a plugin able to propose names for your variables/functions/classes. For instance for the sieve of Eratosthenes:
00:00
Baptiste Rozière
@b_roziere
Jul 10, 2025
Replying to @b_roziere and @SIGKITTEN
I hope we didn't take too long
Upgrading agentic coding capabilities with the new Devstral models | Mistral AI
From mistral.ai
2.9K
Baptiste Rozière
@b_roziere
Jul 9, 2025
Replying to @SIGKITTEN
Good idea. Let me see what we can do.
2.3K
Baptiste Rozière
@b_roziere
Jan 13, 2025
I'm glad to announce that we released an updated version of Codestral today! Codestral-2501 targets completion in IDEs. • twice as fast as the previous version of Codestral 22B • tied for first spot on copilot arena with Deepseek 2.5 (236B total params) and Claude 3.5 sonnet
2.1K
Baptiste Rozière
@b_roziere
Jul 10, 2025
Excited to release Devstral Medium and a new version of Devstral Small! Devstral medium reaches 61.8% on SWE-bench verified, outperforming Gemini 2.5 pro and GPT-4.1 at a fraction of the cost! It is available on our API as `devstral-medium-2507`.
Mistral AI
@MistralAI
Jul 10, 2025
Introducing Devstral Small and Medium 2507! This latest update offers improved performance and cost efficiency, perfectly suited for coding agents and software engineering tasks.
3.4K
Baptiste Rozière
@b_roziere
Aug 24, 2023
Replying to @b_roziere
On top of the base models, we release CodeLlama instruct, which is trained to follow instructions. It is roughly on the same level as Llama 2 34B in terms of general helpfulness abilities, and much better at generating code on a zero-shot version of MBPP.
8.4K
Baptiste Rozière
@b_roziere
Aug 24, 2023
Replying to @b_roziere
Our inference code is available on GitHub: github.com/facebookresear… With @syhw @jnsgehring, @FabianGloeckle, Sten Sootla, @ThomasScialom, @itai_gat, @adiyossLC and others.
2.1K